What actually happened

Since January this year, anyone sending code to the Linux kernel was supposed to add an "Assisted-by" tag if an AI tool had helped write it. According to The Register, the kernel maintainers have now agreed to drop it.

The reasons were refreshingly practical. Not everyone followed the rule anyway. The tag was handing free advertising to the AI companies. And nobody could quite say what the information was for. Not everyone agreed (one maintainer argued that a label sets expectations even if some people ignore it), but the tag is going.

Linus Torvalds, who started Linux and still has the final say, was blunt about it. He doesn't care which tools you used. He expects "people to understand what they are doing."

That's it. That's the whole rule. And it's the one that matters for your software too.

"Was AI used?" is the wrong question

We get asked this a lot now. A business owner is about to hire a developer, or has inherited a system from one, and they want to know whether it was "written by AI". As if the answer tells you whether it's any good.

It doesn't. Almost every developer uses AI in some way now. The latest Stack Overflow survey, reported by The Register on 10 October, had about 30,900 responses. Two thirds of developers use coding assistants or agents, and nearly a third of daily users spend four hours or more of their working day with them. Only 17% said they don't use AI tools at all.

Here's the interesting bit. Only about 7% said they'd trust AI output for important work decisions. Even when the output is easy to check, it's 48%. So the people using it most are the people trusting it least. That's not a contradiction. That's what using a tool properly looks like.

We work the same way. We treat AI like a very fast junior developer: brilliant at typing, happy to have a go at anything, and in need of someone senior to check its work before it goes anywhere near a customer.

The real question: does anyone understand this code?

Torvalds' rule turns a vague worry into something you can actually check. Software that nobody understands is a problem whether a person or a machine wrote it. We've all met the system built by a contractor who vanished, where nobody dares touch the bit that does the invoicing.

AI just makes that situation easier to create, and much faster. You can now have a working app in a weekend without anyone, including you, understanding how it works. It looks finished. It demos beautifully. Then a customer does something unexpected, or someone pokes at your login page, and there's nobody who can explain what's going on underneath.

If you've built something with Lovable, Replit, Bolt or similar, this isn't a telling-off. Those tools are genuinely useful for getting an idea out of your head and in front of people. But the moment real customers, real money or real personal data are involved, someone needs to understand the code. That someone can be you, a developer you hire, or a firm like us. It just can't be nobody.

Questions to ask your developer

Whether you're hiring someone new or checking on what you already have, these are the questions that get to the heart of it. None of them are "did you use AI?"

  • Can you walk me through how this part works? Pick something that matters, like payments or logins. A developer who understands their code can explain it in plain English. One who doesn't will get vague very quickly.
  • What happens when it goes wrong? Ask about a failed payment, a duplicate order, a supplier's system being down. Understanding shows up in the edge cases.
  • Who checks the code before it goes live? AI-written or not, a second pair of eyes catches the daft mistakes. If the answer is "the AI tests it", keep asking.
  • Where is my customers' data, and who can see it? This is where AI-built apps most often come unstuck. If nobody can answer it confidently, that's your top priority.
  • If you disappeared tomorrow, could someone else take over? Understood code tends to be readable code. Code nobody understands tends to be the opposite.

If you get good answers, it genuinely doesn't matter how much AI was involved. If you get blank looks, it doesn't matter how little.

What to do if the answer is "nobody"

It's more common than you'd think, and it's fixable. The usual route is to get someone experienced to read through what you have, tell you honestly what's solid and what isn't, and fix the bits that could hurt you first. Often that's security and data handling, then the things that break when you get more users.

We've written more about hiring a developer to finish an AI-built app, and about the difference between AI-assisted development and vibe coding, if you want to dig in.

The Linux folk have spent decades looking after some of the most important code in the world, and they've landed on a simple rule. Use whatever tools you like. Just understand what you're shipping. We'd say that's a decent rule for a small business app too.

Not sure whether anyone understands your app? Our AI code audit gives you a plain-English answer: what's solid, what's risky and what to fix first. Get in touch and you'll be talking to Rob or Jason, the people who'll actually read your code.

Cheers, Rob